Cancer Diagnosis Support System: Neural Networks Approach

Tumor markers are substances produced by cells of the body in response to cancerous but also to noncancerous conditions. They can be found in body liquids like blood or in tissues and can be used for detection, diagnosis and treatment of some types of cancer. For different types of cancer different tu-mor markers can show abnormal values and the levels of the same tumor marker can be altered in more than one type of cancer. Examples of tumor markers include CA 125 (in ovarian cancer), CA 153 (in breast cancer), CEA (in ovarian, lung, breast, pancreas, and gastrointestinal tract cancers), and PSA (in prostate cancer). Although an abnormal tumor marker level may suggest cancer, tumor markers are not sensitive or specific enough for a reliable cancer diagnosis. But abnormally altered tumor marker values indicate a need for further medical examination. During blood examination only a few tumor marker values are tested and for this reason the usage of such incomplete data for cancer diagnosis support needs estimation of missing marker values. Neural networks are proven tools for prediction tasks on medical data. For example neural networks were applied to differentiate benign from malignant breast conditions base on blood parameters, for diagnosis of different types of liver disease, for early detection of prostate cancer, for studies on blood plasma or for prediction of acute coronary syndromes. In this chapter we present two approaches for a system to support cancer diagnosis. Both systems use heterogeneous neural networks, the first one uses tumor marker values and tumor diagnosis data of thousands of patients for training and testing the neural networks for cancer prediction and the second one extends the tumor marker values by standard blood parameters.